A Variable Length Coding Framework for Cost Function Reduction in Non-Volatile Memory Systems

Abstract: Variable length coding for Non-Volatile Memory (NVM) technologies is a promising method to improve memory capacity and system performance through compressing memory blocks. However, compression techniques used to improve capacity or bandwidth utilization do not take into consideration the asymmetric costs of writing 1's and 0's in NVMs. Taking into account this asymmetry, we propose a variable length encoding framework that reduces the cost of writing data into NVM. Our experimental results on 12 workloads of the SPEC CPU2006 benchmark suite show that, when the cost asymmetry is 1:2, the proposed framework is capable of reducing the NVM programming cost by up to 24% more than leading compression approaches and by 12.5% more than the flip-and-write approach which selects between the data and its complement based on the programming cost.